TensorFlow, the open-source platform for machine learning, has just received its final release for version 2.0. According to the company's official Medium blog, the update brings a set of new features that are meant to boost the performance of the platform, increase its flexibility, and make it more accessible for programmers who are familiar with Python.
TensorFlow 2.0 is driven by the community telling us they want an easy-to-use platform that is both flexible and powerful, and which supports deployment to any platform. TensorFlow 2.0 provides a comprehensive ecosystem of tools for developers, enterprises, and researchers who want to push the state-of-the-art in machine learning and build scalable ML-powered applications.
In TensorFlow 2.0, there are improvements to the low-level API. Redundant APIs have been removed and others have been made more consistent. This will allow users "to build onto the internals of TensorFlow" without having to actually rebuild TensorFlow itself. The latest update also strives to make the platform more accessible and familiar for Python frequents. Tighter integration of the open-source neural network library Keras, and the fact that eager execution is turned on by default should reduce the learning curve for TensorFlow 2.0 and improve debugging efficiency.
The team at TensorFlow has also increased deployment flexibility and options by standardizing the SavedModel file format. This will allow applications to run in a variety of runtimes, such as the web, browser, mobile and embedded systems, and Node.JS.
In performance-oriented scenarios, researchers, casual users, and enthusiasts alike can now utilize the Distribution Strategy API to distribute training across multiple computation hubs. Multi-GPU support is now available on the platform as well as support for custom training loops and distributed training with Model.fit in Keras. Cloud TPU support is on the horizon and will come in a future release.
Moreover, TensorFlow 2.0 also runs three times faster, boasting more performance on graphics processing units:
TensorFlow 2.0 delivers up to 3x faster training performance using mixed precision on Volta and Turing GPUs with a few lines of code, used for example in ResNet-50 and BERT. TensorFlow 2.0 is tightly integrated with TensorRT and uses an improved API to deliver better usability and high performance during inference on NVIDIA T4 Cloud GPUs on Google Cloud.
A video by the TensorFlow channel on YouTube summarizing the changes is as follows:
To help aid users in migrating their projects from TensorFlow 1.x to TensorFlow 2.0, there is a handy guide published here. TensorFlow also has advice on how to be effective and productive with the updated version here. In the same link, you will find the detailed version of all that is new in TensorFlow 2.0.